DCPower-ICS: A Labeled ICS Anomaly Detection Dataset for Data Center Power Infrastructure

DCPower-ICS v1.1.0 is a fully synthetic, uncalibrated labeled dataset for data-center power-control anomaly detection. It contains 24-hour training and test splits sampled at 1 second, with 40 numeric channels. Training includes planned utility interruptions and UPS bypass; test-normal reference periods are grid-connected, while some scripted faults change operating mode. The corrected maintenance ATS transfer is represented in the training data. The test CSV, raw labels and event schedule are preserved. Primary evaluation excludes the unobservable FALSE_HEALTHY_SUBSYSTEM events; equivalent breaker scenarios share a mechanism mapping. Separate normal-only 60-second recovery masks, reference baselines, saved scores and reproducibility analyses are included. Baseline scores are sanity checks, and maintenance false alarms are in-sample diagnostics. The simulator does not provide calibrated electrical-transient dynamics or operational validation. Corresponding source: DCPower-ICS v1.1.0 at https://github.com/vheydari/DCPower-ICS/releases/tag/v1.1.0. The source commit and file hashes are recorded in release_provenance.json and SHA256SUMS.